Color Dynamics, Pigments and Antioxidant Capacity in Pouteria sapota Puree During Frozen Storage: A Correlation Study Using CIELAB Color Space and Machine Learning Models

Abstract

The accurate prediction of bioactive compounds and antioxidant activity in food matrices is critical for optimizing nutritional quality and industrial applications. This study compares the performance of multiple linear regression (MLR) and artificial neural networks (ANN) in predicting antioxidant activity (DPPH, ABTS), total carotenoids, and anthocyanins in mamey pulp, using color variables (CIELab) as predictors. Our results demonstrate that ANN models consistently outperform MLR, achieving lower mean squared error (MSE) and mean absolute error (MAE), alongside higher coefficients of determination (R²). For instance, ANN improved R² values from 0.54 to 0.78 for DPPH, from 0.70 to 0.92 for ABTS, and from 0.45 to 0.87 for total carotenoids. These results highlight the superior ability of ANN to capture nonlinear relationships in complex food systems. Furthermore, the integration of ANN with image analysis techniques offers a promising approach for nondestructive quality control during storage and processing. This research underscores the potential of ANN as a powerful tool for screening bioactive compounds and optimizing functional food development, contributing to advancements in food science and technology.

Keywords: Pouteria sapota; Artificial neural networks; Bioactive compounds; CIELAB color space; Frozen storage; Multiple linear regression model.